What is an authority layer for AI agents?
It is the part that decides what an agent may do on a person's behalf, checks each action before it runs, and writes down the decision.
A person sets the limits. Every call is checked against them.
An agent that books travel, issues refunds or changes a database is acting for someone. An authority layer sits between the agent and the tools it calls. Anything outside the limits does not run, and every decision, yes or no, is recorded.
Who said so?
Each limit traces to a named person's decision.
- Instead of
- A shared service account that nobody chose and nobody hands back.
Is this call inside the limits?
Checked before it runs. No match means no.
- Instead of
- Finding out afterwards, from a log, that the agent went too far.
Can we show what happened?
Allowed and refused calls both leave a record your auditor can check.
- Instead of
- Logs written by the same system that ran the agent.
What you may already have, and what is left over.
An authority layer works beside these. It does not replace them.
Identity and access tools
They prove who a person or an agent is.
- What is left
- What that identity approved for this agent, for this task, for a limited time.
Filters on prompts and answers
They try to spot bad text going in or out.
- What is left
- An authority layer does not read prompts. It limits what the agent can do, whatever the prompt says.
Logging and monitoring
They tell you what happened, after it happened.
- What is left
- A decision made before the call, with the record written by the point that decided.
Three limits, stated plainly.
A call that skips the checkpoint
It is not checked.
- What to do
- Close the other routes: the tool itself can refuse callers that did not come through. See the status on Security and status.
Harm inside the grant
The agent can still misuse what it was given.
- What to do
- Write narrow limits, and hold risky actions for a person to approve.
Compliance
Torvant does not make you compliant with anything.
- What you get
- Records your auditor can check against the frameworks you answer to. Your auditor decides what they are worth.
Torvant is an authority layer. "Power of attorney for AI agents" is how we describe the idea: an analogy, and a grant has no legal force. Nothing is in production use yet. See ten ways agents go wrong and where an authority layer stops, or read the status part by part.
See it on your own agent.
A 30-minute call. We watch one of your agents, block nothing, and show you what your rules would have stopped.